Placental T2* and BOLD effect in response to hyperoxia in normal and growth‐restricted pregnancies: multicenter cohort study

Blood‐oxygen‐level‐dependent (BOLD) magnetic resonance imaging (MRI) facilitates the non‐invasive in‐vivo evaluation of placental oxygenation. The aims of this study were to identify and quantify a relative BOLD effect in response to hyperoxia in the human placenta and to compare it between pregnancies with and those without fetal growth restriction (FGR).

fetal weight (EFW) on ultrasound < 5 th centile) and 75 non-FGR pregnancies (controls) recruited at two centers in Paris, France.Using a 1.5-Tesla MRI system, the same multi-echo gradient-recalled echo (GRE) sequences were performed at both centers to obtain placental T2* values at baseline and in hyperoxic conditions.The relative BOLD effect was calculated according to the equation 100 × (hyperoxic T2* − baseline T2*)/baseline T2*.Baseline T2* values and relative BOLD effect were compared according to EFW (FGR vs non-FGR), presence/absence of Doppler anomalies and birth weight (small-for-gestational age (SGA) vs non-SGA).

INTRODUCTION
Blood-oxygen-level-dependent (BOLD) magnetic resonance imaging (MRI) allows non-invasive in-vivo evaluation of placental oxygenation.Decreased placental oxygenation, mediated by placental insufficiency, is the main mechanism of fetal growth restriction (FGR) 1,2 .BOLD MRI may improve the diagnosis and prediction of FGR, which is, at present, assessed imperfectly by sonographic analysis of fetal growth and uteroplacental Doppler indices.These modalities are unable to directly quantify placental perfusion and lack specificity in differentiating small but healthy fetuses from those affected by placental insufficiency and growth restriction 3 .
BOLD MRI uses hemoglobin as an endogenous contrast agent, whose oxygen saturation varies almost linearly with the transverse relaxation time (T2*) detected on MRI 4 .Because oxyhemoglobin has diamagnetic properties and deoxyhemoglobin is paramagnetic, changes in oxygen saturation of hemoglobin induce changes in local magnetic field susceptibility, thus altering baseline T2*.
The BOLD effect was first demonstrated and validated in a rat model of intrauterine growth restriction 5,6 .It was then demonstrated in the placenta of uncomplicated human pregnancies in 2013 7,8 and, in 2016-2017, also described in cases of FGR [9][10][11] .In 2018, Sinding et al. 12 demonstrated in a population of 49 normal pregnancies and 13 pregnancies complicated by placental dysfunction that the relative change in BOLD signal observed during hyperoxia (BOLD effect) was greater in placentas from complicated pregnancies, but this was solely a reflection of lower baseline T2*.Cases and controls were defined by birth weight (below or at/above the 10 th centile, respectively) and were all considered as non-FGR at the time of MRI.This makes placental T2* a promising marker for distinguishing a well-functioning placenta from a poorly functioning one in vivo 13 , which could be used to improve prenatal diagnosis of FGR 14 .
The objectives of this study were to identify and quantify the BOLD effect in response to hyperoxia in the human placenta and to compare it between non-FGR and FGR pregnancies.Following the findings of recent studies 13,14 , we also wanted to evaluate in our cohort the value of placental T2* as a standalone marker of placental dysfunction.

Subjects and methods
This was a prospective multicenter study (ClinicalTrials .govidentifier: NCT02238301) and was approved by a regional ethics committee (CPP IDF2 2014-06-14).Written informed consent was obtained from all participating women.
Women were recruited from two tertiary university hospitals (H ôpital Necker-Enfants Malades and H ôpital Louis Mourier) in Paris, France, as part of routine prenatal management.Inclusion criteria were singleton pregnancy between 18 and 37 weeks' gestation.Patients with one of the usual contraindications for MRI, those with an abdominal circumference > 125 cm and those with placenta accreta or percreta were excluded.FGR was defined as estimated fetal weight (EFW) < 5 th centile on the last ultrasound scan preceding MRI assessment, with or without abnormal Doppler flow (mean uterine artery (UtA) pulsatility index (PI) > 95 th centile, abnormal umbilical artery (UA) diastolic flow and/or cerebroplacental ratio < 1).Neonatal data were collected and small-for-gestational-age (SGA) neonates were defined as having a birth weight < 10 th centile (hypotrophic).
Values of EFW, birth weight and mean UtA-PI were transformed into Z-scores using previous published formulae [15][16][17] .BOLD results were not communicated to clinicians or patients or used for pregnancy management.

Image acquisition
Data from the last ultrasound scan performed prior to the date of MRI were collected.EFW was calculated using head circumference, abdominal circumference and femur length, according to the Hadlock formula.Biometric and Doppler measurements (UtA, UA and fetal middle cerebral artery) were acquired according to the standardized criteria issued by the French National Committee for Ultrasound Screening 18 .
A 1.5-Tesla MRI system (GE Healthcare, Waukesha, WI, USA) was used in both centers.The pregnant women were placed in a left supine position to reduce aortocaval compression.MRI took place in two stages: (1) placental anatomical MRI followed by baseline placental T2* measurements were obtained in room air with normal oxygen levels (21%); and (2) a non-rebreathing facial mask (Hudson Respiratory Care, Durham, NC, USA) was applied to administer oxygen (100%; 15 L/min for 5 min) and a dynamic BOLD MRI sequence was performed.
Multi-echo gradient-recalled echo (GRE) sequences were performed to obtain T2* values, with identical parameters in both centers: repetition time, 68 ms; 16 echoes ranging from 1.8 ms to 55.1 ms in steps of 3.6 ms; flip angle, 30 • ; slice thickness, 5 mm; interslice spacing, 2 mm; field of view, 400 × 400 mm; and matrix, 256 × 128.The size of the matrix resulted in an in-plane resolution of 1.6 × 1.6 mm.The bandwidth was 244 Hz/pixel.
Image postprocessing was performed using dedicated software (Olea Sphere version 2.4; Olea Medical Software, La Ciotat, France), using the relaxometry module.We therefore obtained quantitative T2* maps (resulting maps) from native sequences (Figure 1).Three methods of slice selection were used to draw the region of interest (ROI): (1) a 'single typical slice', in which a ROI was traced according to the operator's best estimate of the most representative slice of the entire placenta; For all analyses, T2* values of the single typical slice estimated by the first observer were used.The baseline T2* values and BOLD effect to hyperoxia were compared according to EFW (non-FGR vs FGR fetuses), presence/absence of Doppler anomalies and birth weight (SGA vs non-SGA neonates).The correlation of baseline T2* Z-score with gestational age, UtA-PI, EFW and birth weight was also evaluated.

Statistical analysis
The sample size calculation was based on the only published data 19 available at the time of writing the protocol, which was used to make assumptions about signal intensity.From the study of Semple et al. 19 , we expected to observe a mean ± SD signal intensity of 35 ± 15 at baseline and a 20% increase under oxygen therapy with a SD of 20, leading to a mean ± SD difference of 7 ± 14.5, considering a correlation between the two measurements of 0.7.With a two-sided alpha of 5% and a power of 80%, we needed to include 36 patients undergoing MRI before and after oxygen supplementation.Adding a few patients to guarantee sufficient power in case of MRI failure or loss to follow-up, we therefore planned to include 40 women with a non-FGR fetus and the same number of women with a FGR fetus.We also decided to include 20 additional patients with a non-FGR fetus at the start of the protocol to set up the BOLD sequences, test the hyperoxygenation (e.g.flow rate, duration) and select the type of oxygenation mask best tolerated by patients.
Continuous variables were described as median (interquartile range (IQR)) and categorical variables as n (%).Intraclass correlation coefficient (ICC) with 95% CI and Bland-Altman plots were used to examine the agreement between the three methods of slice selection used to obtain baseline T2* values and assess interobserver reproducibility.Wilcoxon signed-rank tests were used to compare baseline and hyperoxic T2* values between patients.Wilcoxon rank-sum tests were used to compare continuous variables and Fisher's exact test was used to compare categorical variables between FGR and non-FGR fetuses and between SGA and non-SGA neonates.Graphical representation and Spearman's correlation coefficient (ρ) were used to assess the relationship of baseline T2* Z-score with gestational age Z-score, UtA-PI Z-score, EFW Z-score and birth weight Z-score.
The statistical software R 4.0.3(www.R-project.org) was used.P-values < 0.05 were considered statistically significant.

RESULTS
Between 18 February 2015 and 26 September 2019, 101 women underwent MRI for a fetal or placental indication and consented to participate in this study, including 68 patients at H ôpital Necker-Enfants Malades and 33 at H ôpital Louis-Mourier, Paris, France.One woman with a multiple gestation was excluded from analysis.Among the 79 non-FGR pregnancies, which constituted the control group, 62 underwent MRI between 31 and < 37 weeks, 11 between 27 and < 31 weeks and six between 18 and < 27 weeks.Among the 21 FGR pregnancies, the respective numbers for the same gestational-age categories were 11, three and seven.The main indication for MRI was brain abnormality, occurring in one-third of cases (n = 31), followed by equivalent rates of urinary, digestive, cardiac or pulmonary abnormalities and FGR.Seven of the 21 FGR fetuses exhibited Doppler abnormality.We encountered image postprocessing difficulties in six examinations, due to maternal or fetal respiratory movements; in all other cases, it was possible to measure a BOLD effect, giving an overall feasibility of 94%.Data for 75 non-FGR and 19 FGR pregnancies were available for statistical analysis (Figure 2).Maternal, imaging and neonatal characteristics are summarized in Table 1.

DISCUSSION
Our study confirms a relative BOLD effect in response to hyperoxia in the human placenta.This is in agreement with previous studies that demonstrated this effect in both uncomplicated human pregnancies 7,8 and cases of FGR with histological signs of maternal hypoperfusion and abnormal Doppler flow 10 .
However, we did not demonstrate a statistically significant difference in the relative BOLD response to hyperoxia between non-FGR and FGR pregnancies, diverging from evidence published previously of a higher relative BOLD effect in the FGR group 12 .This could be due to differences in the approach used to select cases of FGR.In their study, Sinding et al. 12 used a 'postnatal definition' of controls and cases based on birth weight (at/above or below the 10 th centile) and histological examination (exclusion of fetuses with low birth weight and normal placental pathological examination (PPE) and those with normal birth weight and abnormal PPE).This is likely to better capture those cases truly affected by placental dysfunction but relies on information that is not available during prenatal management.Our pragmatic approach used only prenatally available information based on the estimation of fetal weight on ultrasound at the time of MRI and may better reflect the potential of the technique as it would be used in clinical practice.
Conversely, and more importantly, our finding of significantly lower baseline T2* in FGR pregnancies is in
full agreement with previous studies 12,13 .These findings indicate that the increase in the relative BOLD response to hyperoxia observed elsewhere mostly reflects modified baseline conditions.Our study of a large cohort of patients across two centers provides robust data which significantly strengthen the external validity of the results of Sinding et al. [10][11][12] .First, we found a negative correlation between baseline T2* and gestational age, highlighting the ability of T2* to detect physiological senescence of the placenta with advancing gestational age.Similar trends were noted in other large BOLD MRI studies 10,20 , although not in smaller studies 8,9 .Secondly, we found a significant negative association between baseline T2* Z-score and UtA-PI.Elevated UtA-PI is an ultrasound marker of placental insufficiency, reflecting a defect in UtA remodeling.These findings are consistent with those of Sinding et al. 11 .
T2*-weighted MRI has added value in terms of feasibility and cost compared with the more complicated BOLD technique.It reduces the duration of the MRI examination by up to three-fold and eliminates the need for a complex interpretation of the BOLD signal (including confounding factors such as the hemoglobin dissociation curve 21 and interaction with microcirculatory parameters 22 ).However, the potential of the T2* value as a routine marker in pregnancy management remains questionable.Baseline T2* value is certainly a marker of severe placental dysfunction, as seen in FGR fetuses with Doppler abnormality.Our findings confirm those of Sinding et al. 10 but in a larger cohort.However, it remains to be established if T2* baseline values can predict the subsequent occurrence of growth restriction.Our findings in a small number (n = 7) of non-FGR fetuses born SGA do not support this hypothesis, but it should be analyzed in larger cohorts.
Our study shows good agreement between the three methods of slice for computing T2* values.Including more slices or the total placental volume did not improve reproducibility, which is in conflict with the literature.Indeed, the study of Sinding et al. 10 in 2016 showed that the concordance within and between sessions was improved by averaging T2* values from two slices compared with using a single slice.Our study was not dedicated specifically to investigating the various aspects of T2* reproducibility.However, we believe that the addition of more placental slices may favor the inclusion of regions of the placenta with a more variable physiology, or even small non-placental anatomical areas, and increase the heterogeneity of the results and the limits of agreement.
Although our multicenter study confirms the usefulness of T2* in the evaluation of placental function in vivo, we must acknowledge several limitations.Notably, the relatively small sample size stems from challenges with oxygen mask usage and MRI duration.We were not able to recruit as many women with a FGR fetus as we had hoped.Second, Doppler parameters were either missing or not measured on the same day as MRI.However, the interval between MRI and Doppler acquisition was mostly under 8 days and we therefore assumed that their correlation was worth studying and reporting.Third, a recent study suggests that a dysfunctional placenta reaches a hyperoxic BOLD steady state after more than 10 min of oxygen challenge 12 .Given that we used 5 min of oxygen challenge, our results might underestimate the relative BOLD effect in such abnormal placentas.The interpretation of MRI findings must also take into account the presence of subclinical uterine contractions, which were described in the work of Sinding et al. 23 as significantly reducing placental oxygenation.A limitation of our study is that we are, by necessity, assuming that the placenta of a non-FGR fetus with one or more fetal anomalies is a reasonable proxy for a placenta in a completely healthy pregnancy.Similarly, we assumed that the placenta in a FGR pregnancy with or without fetal anomalies is representative of placental function in FGR pregnancies without any anomalies, i.e. pure placental FGR.Until such time as fetal MRI is feasible in large cohorts of truly normal pregnancies, or pregnancies without any abnormality other than ultrasound-estimated SGA, it will be difficult to definitively address this potential concern.However, there is a body of literature that attests that fetal malformations such as those in our cohort are not responsible for placental alterations and, therefore, we believe that this limitation should not be a significant deterrent from further pursuing the study of T2* values as a tool to investigate placental function.We were not able to collect detailed pathology reports of the placentas in our cohort of patients to specifically address this.
In conclusion, this study confirms a BOLD effect in response to hyperoxia in the human placenta.Nevertheless, this effect seems to reflect modified baseline conditions.Consequently, baseline T2* values hold promise as markers of placental dysfunction.Their potential to detect moderate and severe dysfunction before the emergence of FGR necessitates further investigation.

( 2 )
'several slices', by segmentation of the entire placenta with three slices (spaced 15 mm apart) and averaging the three T2* values; and (3) 'all slices', by segmentation of the entire placenta on all slices and averaging all T2* values.For each type of slice selection, ROIs were drawn as the maximum circumference available, taking into account maternal breathing and/or fetal movement.T2* values were obtained by fitting the averaged signal within each ROI as a function of the 16 echo times using the monoexponentially decaying function.To evaluate the interobserver reliability of the single typical slice method, ROIs were delineated by a second independent observer.Therefore, we obtained the following parameters: (1) T2* (ms) in room air, called baseline T2*; (2) T2* (ms) in hyperoxic conditions, called hyperoxic T2*; and (3) relative BOLD effect in response to hyperoxia, calculated as 100 × (hyperoxic T2* − baseline T2*)/baseline T2*.Baseline T2* Z-scores were also computed, using a previous published reference for gestational age 10 , according to the following formulae: Expected baseline T2 * = −4.615× gestational age + 232.278Baseline T2 * Z-score = observed baseline T2 * − expected baseline T2 * /15.83

Figure 1
Figure 1Imaging protocol: from anatomical native sequences (a-f) to functional sequences, before (g-l) and after (m-r) oxygenation, using relaxometry postprocessing module.Images obtained at 28 weeks, 32 weeks and 35 weeks, in fetuses without and those with growth restriction (FGR).

Figure 3
Figure 3 Scatterplots (a,c) and Bland-Altman plots (b,d) showing agreement between three types of slice selection used to obtain baseline T2* values: (a,b) 'all slices' vs 'single typical slice'; and (c,d) 'several slices' vs 'single typical slice'.Mean ( ) and 95% limits of agreement ( ) are shown in (b,d).

Figure 4
Figure 4 Scatterplots (a,c) and Bland-Altman plots (b,d) showing agreement between two observers in baseline placental T2* values (a,b) and relative blood-oxygen-level-dependent (BOLD) response to hyperoxia (c,d).Mean ( ) and 95% limits of agreement ( ) are shown in (b,d).

Table 1
Maternal, imaging and neonatal characteristics of study population, according to presence of fetal growth restriction (FGR) Data are given as median (interquartile range), n/N (%) or n (%).*Non-FGR vs FGR.†Data available for only 67 non-FGR pregnancies and all FGR pregnancies.‡Data available for only six non-FGR pregnancies and 10 FGR pregnancies.§Data available for only 35 non-FGR pregnancies and 16 FGR pregnancies.¶Data available for only 68 non-FGR pregnancies and 17 FGR pregnancies.BOLD, blood-oxygen-level-dependent; EFW, estimated fetal weight; GA, gestational age; MRI, magnetic resonance imaging; PI, pulsatility index; SGA, small-for-gestational age; UA, umbilical artery; UtA, uterine artery.